md-first memory extraction framework for AI agents. Markdown is the single source of truth; SQLite holds state and LanceDB provides the rebuildable vector + BM25 + scalar index. The codebase follows a single-direction DDD layering (entrypoints -> service -> memory -> infra, with component / core / config cross-cutting) enforced by import-linter. Engineering surface: - Coding conventions in .claude/rules/ (path-scoped) and workflows in .claude/skills/ (/commit, /new-branch, /pr). - GitHub Actions CI runs make lint + test + integration; pre-commit mirrors the gates locally (ruff, hygiene hooks, gitlint commit-msg). - Commit messages follow Conventional Commits, enforced by gitlint. - make lint also enforces datetime two-zone discipline and OpenAPI drift.
58 lines
1.7 KiB
Markdown
58 lines
1.7 KiB
Markdown
---
|
|
description: Ask a question about past work. Searches memories and combines with current context to answer.
|
|
arguments:
|
|
- name: question
|
|
description: The question to answer
|
|
required: true
|
|
---
|
|
|
|
# EverMem Ask
|
|
|
|
Answer a question using **both** memory search results **and** current conversation context.
|
|
|
|
## Question
|
|
{{question}}
|
|
|
|
## Instructions
|
|
|
|
1. **Search memories** using `evermem_search` MCP tool with relevant keywords. Start with 10 results.
|
|
|
|
2. **Evaluate results**:
|
|
- If memories provide useful context, note what you learned
|
|
- If more detail needed, search again with different keywords (up to 3 searches)
|
|
- If no relevant memories found, that's OK - proceed with what you know
|
|
|
|
3. **Combine sources** to answer:
|
|
- Memory search results (past sessions)
|
|
- Current conversation context (this session)
|
|
- Your general knowledge (when applicable)
|
|
|
|
4. **Be honest about sources**:
|
|
- "Based on our discussion on [date]..." - when citing memory
|
|
- "From our current session..." - when citing current context
|
|
- "I don't have any recorded information about this" - when memories don't help
|
|
- "Based on general best practices..." - when using general knowledge
|
|
|
|
5. **Admit uncertainty**:
|
|
- If memories are incomplete or unclear, say so
|
|
- If you're inferring rather than recalling, make that clear
|
|
- It's better to say "I don't know" than to guess
|
|
|
|
## Response Format
|
|
|
|
Start with a direct answer, then provide supporting context:
|
|
|
|
```
|
|
[Direct answer to the question]
|
|
|
|
**From memories:**
|
|
- [Relevant points from past sessions, with dates]
|
|
|
|
**Current context:**
|
|
- [Relevant points from this session, if any]
|
|
|
|
**Note:** [Any caveats or gaps in knowledge]
|
|
```
|
|
|
|
Now answer the user's question.
|